The main objective of the project was to provide a corpus for evaluation of various unsupervised techniques of semantic roles annotation. The corpus consist of parts of multilingual parallel corpus PELCRA annotated manually with FrameNet-based semantic roles. PELCRA corpus was chosen as a source corpus in order to be able to evaluate techniques of cross-language projection of semantic roles. This is the also the reason why English frames from the FrameNet project were chosen instead of creating their own version designed for Polish.
Resources
Available resources:
annotated sentences in NEGRA format: srl-20130127.tgz (incomplete unrevised version of 27th January 2013)
Resource statistics:
# of sentences |
266 |
# of annotated frames |
1327 |
# of annotated frame elements |
2518 |